StockVision

Time-series study comparing Prophet and ARIMA forecasts for Amazon and Costco stock prices.

Educational project only. This repository does not provide financial advice or a trading strategy.

StockVision was developed during AI4ALL Ignite (Summer 2025). It focuses on a question that matters more than producing a single forecast: how do two common forecasting approaches behave on the same historical data and evaluation setup?

Results at a glance

Evaluation covers historical AMZN and COST data from January 2022 through June 2025.

Ticker	Model	MAE	RMSE
AMZN	Prophet	37.99	57.01
AMZN	ARIMA	3.26	4.06
COST	Prophet	223.02	236.57
COST	ARIMA	5.34	7.16

Under this project's split, horizon, and configuration, ARIMA produced lower error for both tickers. That result is specific to this experiment; it should not be generalized to other assets, market regimes, horizons, or model configurations.

Visual comparison

Workflow

1. Load and inspect historical price data.
2. Prepare model-specific time-series inputs.
3. Train Prophet and ARIMA baselines.
4. Generate forecasts over the held-out period.
5. Compare MAE and RMSE.
6. Present the outputs in notebooks and a Streamlit interface.

Repository map

MarketForecastAI/
└── New_Stock_Vision/
    ├── Streamlit_app.py        Interactive application
    ├── requirements.txt       Environment snapshot (see setup note)
    ├── notebooks/             Exploration, modeling, and evaluation
    └── data/                  Project datasets

Run the Streamlit app

git clone https://github.com/Muhammad7839/MarketForecastAI.git
cd MarketForecastAI
python -m venv .venv
source .venv/bin/activate      # Windows: .venv\Scripts\activate
pip install streamlit pandas prophet
streamlit run New_Stock_Vision/Streamlit_app.py

Reproducibility notes

• The committed requirements snapshots do not currently declare every dependency used by the Streamlit app. The command above installs the app's direct runtime imports; use an isolated environment and review the notebooks before reproducing the full research workflow.
• Review the notebooks for preprocessing, evaluation, and chart generation.
• Forecast metrics are sensitive to the train/test split and forecast horizon.
• Market prices are non-stationary and affected by events not represented in historical price alone.
• A stronger follow-up would add walk-forward validation, uncertainty calibration, and additional baselines.

Team

• Muhammad A. Imran
• Dieunie Gousse
• Pallavi V.

Developed through AI4ALL Ignite, Summer 2025.